Fabric defect detection is crucial for maintaining the quality and integrity of textile products. Manual inspection methods are time-consuming and subjective, driving the need for automated solutions. The existing model uses a GAN with a dual-discriminator architecture to detect fabric defects. The limitations in this model: increased complexity, longer training times, and higher resource requirements. To address these challenges, an alternative approach is proposed using a single-discriminator GAN, reducing training time and cost while maintaining effectiveness in defect detection. Our system aims to develop a fabric defect detection using improved GAN, this will be reconstructing the fabric images in an unsupervised manner and locate the defect areas by finding differences of original image and the reconstruction. This system consists of the components: generator network and discriminator network. The generator network generates fake images by using Gaussian distribution random noise technique; the discriminator network learns to distinguish between real defect-free fabric images and fake images. The CNN algorithm is applied for training the input images and generate the fake images by finding the parameters like dense, reshape, leakyReLU, Batch Normalization, flatten and upsampling. The performance metrics like accuracy, precision, recall and F1 score are calculated and tested as results are discussed.